Claude Opus 5.5 comes to GitHub Copilot, cheaper and faster

What's new

Anthropic has released its newest AI model, Claude Opus 5.5, to users of GitHub Copilot, the popular AI tool for programmers. The model is gradually rolling out in development environments such as VS Code, Visual Studio, JetBrains, Xcode, Eclipse and Copilot CLI, and it replaces its predecessor Claude Opus 5, which came to Copilot this summer.

What's clever

Opus 5.5 is built for so-called "agentic coding": tasks where the AI works its way through several steps on its own, for example writing code, testing it and fixing errors along the way, without a human having to steer every step. According to Anthropic's and GitHub's own tests, the model solves tasks as well as Opus 5 but uses significantly fewer steps and less computing power, that is, tokens, the pieces of text the AI processes and is paid for. It also recovers faster if something goes wrong during a complex task. In addition, all text from the model is marked with an invisible digital watermark, without costing extra tokens or affecting quality.

Cheaper or better?

The price has dropped noticeably. According to Anthropic's own pricing page, Opus 5.5 costs $4 per million input tokens and $20 per million output tokens via the API, a 20 percent reduction compared with Opus 5. Reuse of text, so-called caching, has become 60 percent cheaper, down to $0.20 per million tokens. Overall, Anthropic expects ordinary tasks to become about 40 percent cheaper to run, because the model both has a lower price and uses fewer tokens per task. There is also a faster "fast mode," up to 2.5 times as fast, at a higher price of $8/$40 per million tokens. On the Terminal-Bench 4.0 benchmark, which tests the AI's ability to solve tasks in a terminal, Opus 5.5 sets a new record according to independent coverage, with a jump of more than 14 points compared with Opus 5. Exactly how it performs against competitors such as OpenAI and Google is not stated directly by GitHub in its announcement.

What it's good at

The model is aimed especially at programmers and companies that use AI to write and fix code, solve multi-step technical tasks and other knowledge work. It is rolling out to subscribers of Copilot Pro+, Max, Business and Enterprise, while users of the cheaper Copilot Pro plan do not get access to the Opus models. Enterprise customers can control via settings in Copilot whether employees should have access to the model.